Fighting Machines

Dan Saxon’s Fighting Machines: Autonomous Weapons and Human Dignity arrived from University of Pennsylvania Press in 2022, at a point when the debate over lethal autonomous weapons had moved out of academic seminars and into the working groups of the United Nations Convention on Certain Conventional Weapons in Geneva. Diplomats had been meeting on the subject since 2014, first informally and then under a Group of Governmental Experts. By the time Saxon’s manuscript went to press, that process had stalled, with a coalition of states pressing for a binding treaty banning fully autonomous weapons, a smaller group of military powers, including the United States, Russia, the United Kingdom and Israel, blocking treaty negotiations, and a middle bloc gesturing toward a political declaration. Into that deadlocked conversation Saxon brought an unusual professional pedigree. He had spent more than a decade as a senior prosecutor at the International Criminal Tribunal for the former Yugoslavia, where he worked on cases including the trial of Slobodan Milošević and led teams investigating crimes against humanity in Kosovo. He left international criminal practice for academia and now teaches international law and human rights at Leiden University College in The Hague. The book reflects both halves of that career: the courtroom prosecutor’s attention to the question of who is responsible when a killing goes wrong, and the academic’s interest in the conceptual frameworks that decide whether killing is permissible at all.

The argument Saxon advances across the book is that human dignity, properly understood, sets a limit on the delegation of lethal decisions to machines. That claim is more demanding than it sounds. Most of the legal literature on autonomous weapons turns on the rules of international humanitarian law that govern targeting in armed conflict: the principle of distinction, which requires combatants to distinguish military objectives from civilians; the principle of proportionality, which forbids attacks expected to cause civilian harm disproportionate to the military advantage anticipated; and the requirement of precautions in attack. Saxon engages with all of those, and he is alert to the question of whether algorithms can perform the contextual judgements those rules demand. But he argues that the discussion needs a second axis. Even an autonomous system that complied perfectly with the targeting rules would, in his view, raise a distinct problem: when a machine selects and engages a human being without a human in the decision loop, the person on the receiving end is killed by a process rather than judged by a peer. The Martens Clause, an old provision of the laws of war first introduced in the preamble to the 1899 Hague Convention and carried forward into Additional Protocol I of the Geneva Conventions, refers to the dictates of the public conscience and the principles of humanity. Saxon reads that clause as more than a rhetorical flourish. He treats it as the doctrinal anchor for a human-dignity test that runs alongside the standard targeting tests and constrains what may be automated.

The book is organised to walk the reader from the legal and ethical foundations through to the operational and policy stakes. Early chapters lay out the relevant bodies of law, principally international humanitarian law, but also international human rights law, which Saxon insists continues to apply in armed conflict and which carries its own protections, particularly around the right to life and the obligation to investigate killings. He then turns to the philosophical literature on human dignity, working through the Kantian tradition that grounds dignity in the rational, autonomous person, the legal articulations in instruments like the Universal Declaration of Human Rights and the German Basic Law, and the way courts, including the German Federal Constitutional Court in its 2006 Aviation Security Act ruling, have used dignity to constrain what governments may do even in extremis. From there he moves to the technology, surveying what current and near-future autonomous systems can and cannot do, before arriving at the policy chapters, which propose a regulatory architecture built around meaningful human control, weapons reviews, accountability mechanisms, and a set of prohibitions on the most decoupled forms of autonomous killing. The closing chapters turn to specific operational settings — urban warfare, peacekeeping, law enforcement — and test the framework against each.

The concrete material Saxon draws on is one of the book’s strengths. He moves through the existing inventory of partly autonomous weapons rather than confining himself to hypothetical futures. The Israeli Harpy and its successor Harop, loitering munitions that detect and attack radar emissions without an in-flight human authorisation for each strike, recur as test cases for what already exists. The Phalanx close-in weapon system mounted on warships, which uses radar to engage incoming missiles in modes where human reaction time would be too slow, illustrates the long-accepted defensive end of the spectrum. The South Korean SGR-A1 sentry gun, deployed along the Demilitarised Zone with the demarcated capability to detect and engage human intruders, sits closer to the line. American programmes such as the X-47B carrier-based unmanned aircraft and successor concepts in unmanned naval and ground systems mark out where the trajectory points. Saxon also looks at how artificial intelligence is moving into the decision-support layer behind the trigger — pattern-of-life analysis for targeting in the United States drone campaigns, the controversies that surrounded Project Maven and its computer-vision work on full-motion video, the use of automated cueing and recommendation systems in air operations. He takes seriously the operational arguments for autonomy: speed of decision in saturation attacks, the protection of one’s own personnel, the ability to operate in denied or contested electromagnetic environments where remote piloting is degraded. He does not dismiss those operational logics. He argues, instead, that they have to be measured against the dignity cost of the resulting kill chains.

Saxon’s treatment of accountability is shaped directly by his prosecutorial background. He devotes substantial attention to the question of who can be held responsible when an autonomous weapon kills unlawfully — the so-called accountability gap that critics have warned about. Working through doctrines of command responsibility, individual criminal responsibility under the Rome Statute, and state responsibility for internationally wrongful acts, he is sceptical of the more alarmist claim that autonomy makes accountability impossible, but he is also sceptical of the more relaxed claim that existing law copes. Programmers, procurement officials, commanders who deploy a system in conditions for which it was not tested, and political leaders who authorise its use all sit in different places along a chain that legal doctrine was not designed for. The book argues for adapting weapons review obligations under Article 36 of Additional Protocol I, which requires states to determine whether a new weapon is lawful before fielding it, into a more rigorous and transparent regime when autonomy is involved. He also examines how internal military doctrine, including United States Department of Defense Directive 3000.09 with its requirement of appropriate levels of human judgement over the use of force, both constrains and leaves room for further autonomy than its language might suggest.

Reception of the book has placed it within an active and crowded field. Paul Scharre’s Army of None, published in 2018 by an author who had worked on the same Defense Department policies from the inside, set much of the agenda; Christof Heyns, the late South African scholar and former United Nations Special Rapporteur on extrajudicial executions, had pressed the human dignity argument in his 2013 report to the Human Rights Council and shaped a generation of follow-on work; the International Committee of the Red Cross, the Campaign to Stop Killer Robots, Human Rights Watch and other civil-society actors have been generating their own reports and statements throughout the period. Saxon’s contribution is read in that context as the most sustained legal-academic effort to put human dignity, rather than targeting compliance, at the centre of the case for regulation. Reviewers have generally welcomed the doctrinal rigour and the courtroom-tested clarity about responsibility, while pushing back in two main directions. The first counterpoint, voiced by writers more sympathetic to defence ministries and by some operational lawyers, is that Saxon’s dignity-based limit is doing more work than the law as it stands actually supports — that the Martens Clause is too thin a doctrinal reed to bear the weight of a ban, and that proportionality and distinction already supply the relevant constraints. The second, from the more abolitionist end, is the opposite charge: that by accepting a regulatory architecture rather than a categorical prohibition, the book concedes too much ground to states that will then carve out wide exceptions for defensive systems, anti-materiel weapons and counter-drone applications. Both critiques are real, and the book in some places anticipates them, in other places leaves them open.

For a reader assembling a working library on artificial intelligence in war, Fighting Machines occupies a specific position. It is not the operational survey that Scharre’s books provide, nor the engineering account that one finds in technical literature, nor the campaign manifesto that the advocacy reports offer. It is closest in genre to academic legal monographs by scholars such as Rebecca Crootof, Kenneth Anderson and Matthew Waxman, Michael Schmitt, and Marco Sassòli, but distinguished by the dignity frame and by the prosecutor’s eye for the chain of responsibility. It pairs naturally with Heyns’s reports for the human rights anchor, with the ICRC’s position papers for the humanitarian law anchor, and with the more recent operational and policy literature for the empirical anchor. Readers coming from a policy or diplomatic background will find a usable framework for the Geneva discussions; readers coming from philosophy will find a legal scaffolding for what otherwise remain abstract dignity claims; readers coming from the defence community will find an argument that has to be answered rather than a polemic that can be set aside. Where the book does less is on the strategic and geopolitical layer — the implications of autonomy for deterrence, alliance politics, the arms race dynamics between the United States, China and Russia, the way autonomy is reshaping the war in Ukraine and the security posture of European militaries that have suddenly discovered they need a defence industrial base. Those questions sit largely outside its frame.

Looking back from a few years on, the durable parts of Fighting Machines are likely to be the doctrinal architecture and the accountability analysis. The Martens Clause reading, the application of Article 36 reviews to autonomous systems, the mapping of criminal and state responsibility onto distributed kill chains — those are arguments that will be cited as the regulatory conversation moves forward, whether or not the specific policy proposals are adopted. The parts most exposed to ageing are the technology survey, where the inventory of systems and capabilities has already shifted under the pressure of operational experience in Ukraine and elsewhere, and the diplomatic chapter, written before the most recent rounds of the CCW Group of Governmental Experts and before the broader push by the United Nations General Assembly that began producing its own resolutions on autonomous weapons. Saxon’s book closed one phase of the debate, in which the field worked out the conceptual vocabulary and the legal questions; the phase that has opened since is operational and geopolitical, and will be written largely by other hands. What Fighting Machines leaves behind for that next phase is a hard, specific claim — that a human being is owed a human decision when their life is at stake — and a worked-out apparatus for testing how far that claim can be pushed inside the laws of war.